kernel/fs/ocfs2/cluster/nodemanager.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public
  17  * License along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA.
  20  */
  21
  22 #include <linux/slab.h>
  23 #include <linux/kernel.h>
  24 #include <linux/module.h>
  25 #include <linux/configfs.h>
  26
  27 #include "tcp.h"
  28 #include "nodemanager.h"
  29 #include "heartbeat.h"
  30 #include "masklog.h"
  31 #include "sys.h"
  32
  33 /* for now we operate under the assertion that there can be only one
  34  * cluster active at a time.  Changing this will require trickling
  35  * cluster references throughout where nodes are looked up */
  36 struct o2nm_cluster *o2nm_single_cluster = NULL;
  37
  38 char *o2nm_fence_method_desc[O2NM_FENCE_METHODS] = {
  39                 "reset",        /* O2NM_FENCE_RESET */
  40                 "panic",        /* O2NM_FENCE_PANIC */
  41 };
  42
  43 struct o2nm_node *o2nm_get_node_by_num(u8 node_num)
  44 {
  45         struct o2nm_node *node = NULL;
  46
  47         if (node_num >= O2NM_MAX_NODES || o2nm_single_cluster == NULL)
  48                 goto out;
  49
  50         read_lock(&o2nm_single_cluster->cl_nodes_lock);
  51         node = o2nm_single_cluster->cl_nodes[node_num];
  52         if (node)
  53                 config_item_get(&node->nd_item);
  54         read_unlock(&o2nm_single_cluster->cl_nodes_lock);
  55 out:
  56         return node;
  57 }
  58 EXPORT_SYMBOL_GPL(o2nm_get_node_by_num);
  59
  60 int o2nm_configured_node_map(unsigned long *map, unsigned bytes)
  61 {
  62         struct o2nm_cluster *cluster = o2nm_single_cluster;
  63
  64         BUG_ON(bytes < (sizeof(cluster->cl_nodes_bitmap)));
  65
  66         if (cluster == NULL)
  67                 return -EINVAL;
  68
  69         read_lock(&cluster->cl_nodes_lock);
  70         memcpy(map, cluster->cl_nodes_bitmap, sizeof(cluster->cl_nodes_bitmap));
  71         read_unlock(&cluster->cl_nodes_lock);
  72
  73         return 0;
  74 }
  75 EXPORT_SYMBOL_GPL(o2nm_configured_node_map);
  76
  77 static struct o2nm_node *o2nm_node_ip_tree_lookup(struct o2nm_cluster *cluster,
  78                                                   __be32 ip_needle,
  79                                                   struct rb_node ***ret_p,
  80                                                   struct rb_node **ret_parent)
  81 {
  82         struct rb_node **p = &cluster->cl_node_ip_tree.rb_node;
  83         struct rb_node *parent = NULL;
  84         struct o2nm_node *node, *ret = NULL;
  85
  86         while (*p) {
  87                 int cmp;
  88
  89                 parent = *p;
  90                 node = rb_entry(parent, struct o2nm_node, nd_ip_node);
  91
  92                 cmp = memcmp(&ip_needle, &node->nd_ipv4_address,
  93                                 sizeof(ip_needle));
  94                 if (cmp < 0)
  95                         p = &(*p)->rb_left;
  96                 else if (cmp > 0)
  97                         p = &(*p)->rb_right;
  98                 else {
  99                         ret = node;
 100                         break;
 101                 }
 102         }
 103
 104         if (ret_p != NULL)
 105                 *ret_p = p;
 106         if (ret_parent != NULL)
 107                 *ret_parent = parent;
 108
 109         return ret;
 110 }
 111
 112 struct o2nm_node *o2nm_get_node_by_ip(__be32 addr)
 113 {
 114         struct o2nm_node *node = NULL;
 115         struct o2nm_cluster *cluster = o2nm_single_cluster;
 116
 117         if (cluster == NULL)
 118                 goto out;
 119
 120         read_lock(&cluster->cl_nodes_lock);
 121         node = o2nm_node_ip_tree_lookup(cluster, addr, NULL, NULL);
 122         if (node)
 123                 config_item_get(&node->nd_item);
 124         read_unlock(&cluster->cl_nodes_lock);
 125
 126 out:
 127         return node;
 128 }
 129 EXPORT_SYMBOL_GPL(o2nm_get_node_by_ip);
 130
 131 void o2nm_node_put(struct o2nm_node *node)
 132 {
 133         config_item_put(&node->nd_item);
 134 }
 135 EXPORT_SYMBOL_GPL(o2nm_node_put);
 136
 137 void o2nm_node_get(struct o2nm_node *node)
 138 {
 139         config_item_get(&node->nd_item);
 140 }
 141 EXPORT_SYMBOL_GPL(o2nm_node_get);
 142
 143 u8 o2nm_this_node(void)
 144 {
 145         u8 node_num = O2NM_MAX_NODES;
 146
 147         if (o2nm_single_cluster && o2nm_single_cluster->cl_has_local)
 148                 node_num = o2nm_single_cluster->cl_local_node;
 149
 150         return node_num;
 151 }
 152 EXPORT_SYMBOL_GPL(o2nm_this_node);
 153
 154 /* node configfs bits */
 155
 156 static struct o2nm_cluster *to_o2nm_cluster(struct config_item *item)
 157 {
 158         return item ?
 159                 container_of(to_config_group(item), struct o2nm_cluster,
 160                              cl_group)
 161                 : NULL;
 162 }
 163
 164 static struct o2nm_node *to_o2nm_node(struct config_item *item)
 165 {
 166         return item ? container_of(item, struct o2nm_node, nd_item) : NULL;
 167 }
 168
 169 static void o2nm_node_release(struct config_item *item)
 170 {
 171         struct o2nm_node *node = to_o2nm_node(item);
 172         kfree(node);
 173 }
 174
 175 static ssize_t o2nm_node_num_show(struct config_item *item, char *page)
 176 {
 177         return sprintf(page, "%d\n", to_o2nm_node(item)->nd_num);
 178 }
 179
 180 static struct o2nm_cluster *to_o2nm_cluster_from_node(struct o2nm_node *node)
 181 {
 182         /* through the first node_set .parent
 183          * mycluster/nodes/mynode == o2nm_cluster->o2nm_node_group->o2nm_node */
 184         return to_o2nm_cluster(node->nd_item.ci_parent->ci_parent);
 185 }
 186
 187 enum {
 188         O2NM_NODE_ATTR_NUM = 0,
 189         O2NM_NODE_ATTR_PORT,
 190         O2NM_NODE_ATTR_ADDRESS,
 191 };
 192
 193 static ssize_t o2nm_node_num_store(struct config_item *item, const char *page,
 194                                    size_t count)
 195 {
 196         struct o2nm_node *node = to_o2nm_node(item);
 197         struct o2nm_cluster *cluster = to_o2nm_cluster_from_node(node);
 198         unsigned long tmp;
 199         char *p = (char *)page;
 200         int ret = 0;
 201
 202         tmp = simple_strtoul(p, &p, 0);
 203         if (!p || (*p && (*p != '\n')))
 204                 return -EINVAL;
 205
 206         if (tmp >= O2NM_MAX_NODES)
 207                 return -ERANGE;
 208
 209         /* once we're in the cl_nodes tree networking can look us up by
 210          * node number and try to use our address and port attributes
 211          * to connect to this node.. make sure that they've been set
 212          * before writing the node attribute? */
 213         if (!test_bit(O2NM_NODE_ATTR_ADDRESS, &node->nd_set_attributes) ||
 214             !test_bit(O2NM_NODE_ATTR_PORT, &node->nd_set_attributes))
 215                 return -EINVAL; /* XXX */
 216
 217         write_lock(&cluster->cl_nodes_lock);
 218         if (cluster->cl_nodes[tmp])
 219                 ret = -EEXIST;
 220         else if (test_and_set_bit(O2NM_NODE_ATTR_NUM,
 221                         &node->nd_set_attributes))
 222                 ret = -EBUSY;
 223         else  {
 224                 cluster->cl_nodes[tmp] = node;
 225                 node->nd_num = tmp;
 226                 set_bit(tmp, cluster->cl_nodes_bitmap);
 227         }
 228         write_unlock(&cluster->cl_nodes_lock);
 229         if (ret)
 230                 return ret;
 231
 232         return count;
 233 }
 234 static ssize_t o2nm_node_ipv4_port_show(struct config_item *item, char *page)
 235 {
 236         return sprintf(page, "%u\n", ntohs(to_o2nm_node(item)->nd_ipv4_port));
 237 }
 238
 239 static ssize_t o2nm_node_ipv4_port_store(struct config_item *item,
 240                                          const char *page, size_t count)
 241 {
 242         struct o2nm_node *node = to_o2nm_node(item);
 243         unsigned long tmp;
 244         char *p = (char *)page;
 245
 246         tmp = simple_strtoul(p, &p, 0);
 247         if (!p || (*p && (*p != '\n')))
 248                 return -EINVAL;
 249
 250         if (tmp == 0)
 251                 return -EINVAL;
 252         if (tmp >= (u16)-1)
 253                 return -ERANGE;
 254
 255         if (test_and_set_bit(O2NM_NODE_ATTR_PORT, &node->nd_set_attributes))
 256                 return -EBUSY;
 257         node->nd_ipv4_port = htons(tmp);
 258
 259         return count;
 260 }
 261
 262 static ssize_t o2nm_node_ipv4_address_show(struct config_item *item, char *page)
 263 {
 264         return sprintf(page, "%pI4\n", &to_o2nm_node(item)->nd_ipv4_address);
 265 }
 266
 267 static ssize_t o2nm_node_ipv4_address_store(struct config_item *item,
 268                                             const char *page,
 269                                             size_t count)
 270 {
 271         struct o2nm_node *node = to_o2nm_node(item);
 272         struct o2nm_cluster *cluster = to_o2nm_cluster_from_node(node);
 273         int ret, i;
 274         struct rb_node **p, *parent;
 275         unsigned int octets[4];
 276         __be32 ipv4_addr = 0;
 277
 278         ret = sscanf(page, "%3u.%3u.%3u.%3u", &octets[3], &octets[2],
 279                      &octets[1], &octets[0]);
 280         if (ret != 4)
 281                 return -EINVAL;
 282
 283         for (i = 0; i < ARRAY_SIZE(octets); i++) {
 284                 if (octets[i] > 255)
 285                         return -ERANGE;
 286                 be32_add_cpu(&ipv4_addr, octets[i] << (i * 8));
 287         }
 288
 289         ret = 0;
 290         write_lock(&cluster->cl_nodes_lock);
 291         if (o2nm_node_ip_tree_lookup(cluster, ipv4_addr, &p, &parent))
 292                 ret = -EEXIST;
 293         else if (test_and_set_bit(O2NM_NODE_ATTR_ADDRESS,
 294                         &node->nd_set_attributes))
 295                 ret = -EBUSY;
 296         else {
 297                 rb_link_node(&node->nd_ip_node, parent, p);
 298                 rb_insert_color(&node->nd_ip_node, &cluster->cl_node_ip_tree);
 299         }
 300         write_unlock(&cluster->cl_nodes_lock);
 301         if (ret)
 302                 return ret;
 303
 304         memcpy(&node->nd_ipv4_address, &ipv4_addr, sizeof(ipv4_addr));
 305
 306         return count;
 307 }
 308
 309 static ssize_t o2nm_node_local_show(struct config_item *item, char *page)
 310 {
 311         return sprintf(page, "%d\n", to_o2nm_node(item)->nd_local);
 312 }
 313
 314 static ssize_t o2nm_node_local_store(struct config_item *item, const char *page,
 315                                      size_t count)
 316 {
 317         struct o2nm_node *node = to_o2nm_node(item);
 318         struct o2nm_cluster *cluster = to_o2nm_cluster_from_node(node);
 319         unsigned long tmp;
 320         char *p = (char *)page;
 321         ssize_t ret;
 322
 323         tmp = simple_strtoul(p, &p, 0);
 324         if (!p || (*p && (*p != '\n')))
 325                 return -EINVAL;
 326
 327         tmp = !!tmp; /* boolean of whether this node wants to be local */
 328
 329         /* setting local turns on networking rx for now so we require having
 330          * set everything else first */
 331         if (!test_bit(O2NM_NODE_ATTR_ADDRESS, &node->nd_set_attributes) ||
 332             !test_bit(O2NM_NODE_ATTR_NUM, &node->nd_set_attributes) ||
 333             !test_bit(O2NM_NODE_ATTR_PORT, &node->nd_set_attributes))
 334                 return -EINVAL; /* XXX */
 335
 336         /* the only failure case is trying to set a new local node
 337          * when a different one is already set */
 338         if (tmp && tmp == cluster->cl_has_local &&
 339             cluster->cl_local_node != node->nd_num)
 340                 return -EBUSY;
 341
 342         /* bring up the rx thread if we're setting the new local node. */
 343         if (tmp && !cluster->cl_has_local) {
 344                 ret = o2net_start_listening(node);
 345                 if (ret)
 346                         return ret;
 347         }
 348
 349         if (!tmp && cluster->cl_has_local &&
 350             cluster->cl_local_node == node->nd_num) {
 351                 o2net_stop_listening(node);
 352                 cluster->cl_local_node = O2NM_INVALID_NODE_NUM;
 353         }
 354
 355         node->nd_local = tmp;
 356         if (node->nd_local) {
 357                 cluster->cl_has_local = tmp;
 358                 cluster->cl_local_node = node->nd_num;
 359         }
 360
 361         return count;
 362 }
 363
 364 CONFIGFS_ATTR(o2nm_node_, num);
 365 CONFIGFS_ATTR(o2nm_node_, ipv4_port);
 366 CONFIGFS_ATTR(o2nm_node_, ipv4_address);
 367 CONFIGFS_ATTR(o2nm_node_, local);
 368
 369 static struct configfs_attribute *o2nm_node_attrs[] = {
 370         &o2nm_node_attr_num,
 371         &o2nm_node_attr_ipv4_port,
 372         &o2nm_node_attr_ipv4_address,
 373         &o2nm_node_attr_local,
 374         NULL,
 375 };
 376
 377 static struct configfs_item_operations o2nm_node_item_ops = {
 378         .release                = o2nm_node_release,
 379 };
 380
 381 static struct config_item_type o2nm_node_type = {
 382         .ct_item_ops    = &o2nm_node_item_ops,
 383         .ct_attrs       = o2nm_node_attrs,
 384         .ct_owner       = THIS_MODULE,
 385 };
 386
 387 /* node set */
 388
 389 struct o2nm_node_group {
 390         struct config_group ns_group;
 391         /* some stuff? */
 392 };
 393
 394 #if 0
 395 static struct o2nm_node_group *to_o2nm_node_group(struct config_group *group)
 396 {
 397         return group ?
 398                 container_of(group, struct o2nm_node_group, ns_group)
 399                 : NULL;
 400 }
 401 #endif
 402
 403 static ssize_t o2nm_cluster_attr_write(const char *page, ssize_t count,
 404                                        unsigned int *val)
 405 {
 406         unsigned long tmp;
 407         char *p = (char *)page;
 408
 409         tmp = simple_strtoul(p, &p, 0);
 410         if (!p || (*p && (*p != '\n')))
 411                 return -EINVAL;
 412
 413         if (tmp == 0)
 414                 return -EINVAL;
 415         if (tmp >= (u32)-1)
 416                 return -ERANGE;
 417
 418         *val = tmp;
 419
 420         return count;
 421 }
 422
 423 static ssize_t o2nm_cluster_idle_timeout_ms_show(struct config_item *item,
 424         char *page)
 425 {
 426         return sprintf(page, "%u\n", to_o2nm_cluster(item)->cl_idle_timeout_ms);
 427 }
 428
 429 static ssize_t o2nm_cluster_idle_timeout_ms_store(struct config_item *item,
 430         const char *page, size_t count)
 431 {
 432         struct o2nm_cluster *cluster = to_o2nm_cluster(item);
 433         ssize_t ret;
 434         unsigned int val;
 435
 436         ret =  o2nm_cluster_attr_write(page, count, &val);
 437
 438         if (ret > 0) {
 439                 if (cluster->cl_idle_timeout_ms != val
 440                         && o2net_num_connected_peers()) {
 441                         mlog(ML_NOTICE,
 442                              "o2net: cannot change idle timeout after "
 443                              "the first peer has agreed to it."
 444                              "  %d connected peers\n",
 445                              o2net_num_connected_peers());
 446                         ret = -EINVAL;
 447                 } else if (val <= cluster->cl_keepalive_delay_ms) {
 448                         mlog(ML_NOTICE, "o2net: idle timeout must be larger "
 449                              "than keepalive delay\n");
 450                         ret = -EINVAL;
 451                 } else {
 452                         cluster->cl_idle_timeout_ms = val;
 453                 }
 454         }
 455
 456         return ret;
 457 }
 458
 459 static ssize_t o2nm_cluster_keepalive_delay_ms_show(
 460         struct config_item *item, char *page)
 461 {
 462         return sprintf(page, "%u\n",
 463                         to_o2nm_cluster(item)->cl_keepalive_delay_ms);
 464 }
 465
 466 static ssize_t o2nm_cluster_keepalive_delay_ms_store(
 467         struct config_item *item, const char *page, size_t count)
 468 {
 469         struct o2nm_cluster *cluster = to_o2nm_cluster(item);
 470         ssize_t ret;
 471         unsigned int val;
 472
 473         ret =  o2nm_cluster_attr_write(page, count, &val);
 474
 475         if (ret > 0) {
 476                 if (cluster->cl_keepalive_delay_ms != val
 477                     && o2net_num_connected_peers()) {
 478                         mlog(ML_NOTICE,
 479                              "o2net: cannot change keepalive delay after"
 480                              " the first peer has agreed to it."
 481                              "  %d connected peers\n",
 482                              o2net_num_connected_peers());
 483                         ret = -EINVAL;
 484                 } else if (val >= cluster->cl_idle_timeout_ms) {
 485                         mlog(ML_NOTICE, "o2net: keepalive delay must be "
 486                              "smaller than idle timeout\n");
 487                         ret = -EINVAL;
 488                 } else {
 489                         cluster->cl_keepalive_delay_ms = val;
 490                 }
 491         }
 492
 493         return ret;
 494 }
 495
 496 static ssize_t o2nm_cluster_reconnect_delay_ms_show(
 497         struct config_item *item, char *page)
 498 {
 499         return sprintf(page, "%u\n",
 500                         to_o2nm_cluster(item)->cl_reconnect_delay_ms);
 501 }
 502
 503 static ssize_t o2nm_cluster_reconnect_delay_ms_store(
 504         struct config_item *item, const char *page, size_t count)
 505 {
 506         return o2nm_cluster_attr_write(page, count,
 507                                &to_o2nm_cluster(item)->cl_reconnect_delay_ms);
 508 }
 509
 510 static ssize_t o2nm_cluster_fence_method_show(
 511         struct config_item *item, char *page)
 512 {
 513         struct o2nm_cluster *cluster = to_o2nm_cluster(item);
 514         ssize_t ret = 0;
 515
 516         if (cluster)
 517                 ret = sprintf(page, "%s\n",
 518                               o2nm_fence_method_desc[cluster->cl_fence_method]);
 519         return ret;
 520 }
 521
 522 static ssize_t o2nm_cluster_fence_method_store(
 523         struct config_item *item, const char *page, size_t count)
 524 {
 525         unsigned int i;
 526
 527         if (page[count - 1] != '\n')
 528                 goto bail;
 529
 530         for (i = 0; i < O2NM_FENCE_METHODS; ++i) {
 531                 if (count != strlen(o2nm_fence_method_desc[i]) + 1)
 532                         continue;
 533                 if (strncasecmp(page, o2nm_fence_method_desc[i], count - 1))
 534                         continue;
 535                 if (to_o2nm_cluster(item)->cl_fence_method != i) {
 536                         printk(KERN_INFO "ocfs2: Changing fence method to %s\n",
 537                                o2nm_fence_method_desc[i]);
 538                         to_o2nm_cluster(item)->cl_fence_method = i;
 539                 }
 540                 return count;
 541         }
 542
 543 bail:
 544         return -EINVAL;
 545 }
 546
 547 CONFIGFS_ATTR(o2nm_cluster_, idle_timeout_ms);
 548 CONFIGFS_ATTR(o2nm_cluster_, keepalive_delay_ms);
 549 CONFIGFS_ATTR(o2nm_cluster_, reconnect_delay_ms);
 550 CONFIGFS_ATTR(o2nm_cluster_, fence_method);
 551
 552 static struct configfs_attribute *o2nm_cluster_attrs[] = {
 553         &o2nm_cluster_attr_idle_timeout_ms,
 554         &o2nm_cluster_attr_keepalive_delay_ms,
 555         &o2nm_cluster_attr_reconnect_delay_ms,
 556         &o2nm_cluster_attr_fence_method,
 557         NULL,
 558 };
 559
 560 static struct config_item *o2nm_node_group_make_item(struct config_group *group,
 561                                                      const char *name)
 562 {
 563         struct o2nm_node *node = NULL;
 564
 565         if (strlen(name) > O2NM_MAX_NAME_LEN)
 566                 return ERR_PTR(-ENAMETOOLONG);
 567
 568         node = kzalloc(sizeof(struct o2nm_node), GFP_KERNEL);
 569         if (node == NULL)
 570                 return ERR_PTR(-ENOMEM);
 571
 572         strcpy(node->nd_name, name); /* use item.ci_namebuf instead? */
 573         config_item_init_type_name(&node->nd_item, name, &o2nm_node_type);
 574         spin_lock_init(&node->nd_lock);
 575
 576         mlog(ML_CLUSTER, "o2nm: Registering node %s\n", name);
 577
 578         return &node->nd_item;
 579 }
 580
 581 static void o2nm_node_group_drop_item(struct config_group *group,
 582                                       struct config_item *item)
 583 {
 584         struct o2nm_node *node = to_o2nm_node(item);
 585         struct o2nm_cluster *cluster = to_o2nm_cluster(group->cg_item.ci_parent);
 586
 587         o2net_disconnect_node(node);
 588
 589         if (cluster->cl_has_local &&
 590             (cluster->cl_local_node == node->nd_num)) {
 591                 cluster->cl_has_local = 0;
 592                 cluster->cl_local_node = O2NM_INVALID_NODE_NUM;
 593                 o2net_stop_listening(node);
 594         }
 595
 596         /* XXX call into net to stop this node from trading messages */
 597
 598         write_lock(&cluster->cl_nodes_lock);
 599
 600         /* XXX sloppy */
 601         if (node->nd_ipv4_address)
 602                 rb_erase(&node->nd_ip_node, &cluster->cl_node_ip_tree);
 603
 604         /* nd_num might be 0 if the node number hasn't been set.. */
 605         if (cluster->cl_nodes[node->nd_num] == node) {
 606                 cluster->cl_nodes[node->nd_num] = NULL;
 607                 clear_bit(node->nd_num, cluster->cl_nodes_bitmap);
 608         }
 609         write_unlock(&cluster->cl_nodes_lock);
 610
 611         mlog(ML_CLUSTER, "o2nm: Unregistered node %s\n",
 612              config_item_name(&node->nd_item));
 613
 614         config_item_put(item);
 615 }
 616
 617 static struct configfs_group_operations o2nm_node_group_group_ops = {
 618         .make_item      = o2nm_node_group_make_item,
 619         .drop_item      = o2nm_node_group_drop_item,
 620 };
 621
 622 static struct config_item_type o2nm_node_group_type = {
 623         .ct_group_ops   = &o2nm_node_group_group_ops,
 624         .ct_owner       = THIS_MODULE,
 625 };
 626
 627 /* cluster */
 628
 629 static void o2nm_cluster_release(struct config_item *item)
 630 {
 631         struct o2nm_cluster *cluster = to_o2nm_cluster(item);
 632
 633         kfree(cluster->cl_group.default_groups);
 634         kfree(cluster);
 635 }
 636
 637 static struct configfs_item_operations o2nm_cluster_item_ops = {
 638         .release        = o2nm_cluster_release,
 639 };
 640
 641 static struct config_item_type o2nm_cluster_type = {
 642         .ct_item_ops    = &o2nm_cluster_item_ops,
 643         .ct_attrs       = o2nm_cluster_attrs,
 644         .ct_owner       = THIS_MODULE,
 645 };
 646
 647 /* cluster set */
 648
 649 struct o2nm_cluster_group {
 650         struct configfs_subsystem cs_subsys;
 651         /* some stuff? */
 652 };
 653
 654 #if 0
 655 static struct o2nm_cluster_group *to_o2nm_cluster_group(struct config_group *group)
 656 {
 657         return group ?
 658                 container_of(to_configfs_subsystem(group), struct o2nm_cluster_group, cs_subsys)
 659                : NULL;
 660 }
 661 #endif
 662
 663 static struct config_group *o2nm_cluster_group_make_group(struct config_group *group,
 664                                                           const char *name)
 665 {
 666         struct o2nm_cluster *cluster = NULL;
 667         struct o2nm_node_group *ns = NULL;
 668         struct config_group *o2hb_group = NULL, *ret = NULL;
 669         void *defs = NULL;
 670
 671         /* this runs under the parent dir's i_mutex; there can be only
 672          * one caller in here at a time */
 673         if (o2nm_single_cluster)
 674                 return ERR_PTR(-ENOSPC);
 675
 676         cluster = kzalloc(sizeof(struct o2nm_cluster), GFP_KERNEL);
 677         ns = kzalloc(sizeof(struct o2nm_node_group), GFP_KERNEL);
 678         defs = kcalloc(3, sizeof(struct config_group *), GFP_KERNEL);
 679         o2hb_group = o2hb_alloc_hb_set();
 680         if (cluster == NULL || ns == NULL || o2hb_group == NULL || defs == NULL)
 681                 goto out;
 682
 683         config_group_init_type_name(&cluster->cl_group, name,
 684                                     &o2nm_cluster_type);
 685         config_group_init_type_name(&ns->ns_group, "node",
 686                                     &o2nm_node_group_type);
 687
 688         cluster->cl_group.default_groups = defs;
 689         cluster->cl_group.default_groups[0] = &ns->ns_group;
 690         cluster->cl_group.default_groups[1] = o2hb_group;
 691         cluster->cl_group.default_groups[2] = NULL;
 692         rwlock_init(&cluster->cl_nodes_lock);
 693         cluster->cl_node_ip_tree = RB_ROOT;
 694         cluster->cl_reconnect_delay_ms = O2NET_RECONNECT_DELAY_MS_DEFAULT;
 695         cluster->cl_idle_timeout_ms    = O2NET_IDLE_TIMEOUT_MS_DEFAULT;
 696         cluster->cl_keepalive_delay_ms = O2NET_KEEPALIVE_DELAY_MS_DEFAULT;
 697         cluster->cl_fence_method       = O2NM_FENCE_RESET;
 698
 699         ret = &cluster->cl_group;
 700         o2nm_single_cluster = cluster;
 701
 702 out:
 703         if (ret == NULL) {
 704                 kfree(cluster);
 705                 kfree(ns);
 706                 o2hb_free_hb_set(o2hb_group);
 707                 kfree(defs);
 708                 ret = ERR_PTR(-ENOMEM);
 709         }
 710
 711         return ret;
 712 }
 713
 714 static void o2nm_cluster_group_drop_item(struct config_group *group, struct config_item *item)
 715 {
 716         struct o2nm_cluster *cluster = to_o2nm_cluster(item);
 717         int i;
 718         struct config_item *killme;
 719
 720         BUG_ON(o2nm_single_cluster != cluster);
 721         o2nm_single_cluster = NULL;
 722
 723         for (i = 0; cluster->cl_group.default_groups[i]; i++) {
 724                 killme = &cluster->cl_group.default_groups[i]->cg_item;
 725                 cluster->cl_group.default_groups[i] = NULL;
 726                 config_item_put(killme);
 727         }
 728
 729         config_item_put(item);
 730 }
 731
 732 static struct configfs_group_operations o2nm_cluster_group_group_ops = {
 733         .make_group     = o2nm_cluster_group_make_group,
 734         .drop_item      = o2nm_cluster_group_drop_item,
 735 };
 736
 737 static struct config_item_type o2nm_cluster_group_type = {
 738         .ct_group_ops   = &o2nm_cluster_group_group_ops,
 739         .ct_owner       = THIS_MODULE,
 740 };
 741
 742 static struct o2nm_cluster_group o2nm_cluster_group = {
 743         .cs_subsys = {
 744                 .su_group = {
 745                         .cg_item = {
 746                                 .ci_namebuf = "cluster",
 747                                 .ci_type = &o2nm_cluster_group_type,
 748                         },
 749                 },
 750         },
 751 };
 752
 753 int o2nm_depend_item(struct config_item *item)
 754 {
 755         return configfs_depend_item(&o2nm_cluster_group.cs_subsys, item);
 756 }
 757
 758 void o2nm_undepend_item(struct config_item *item)
 759 {
 760         configfs_undepend_item(&o2nm_cluster_group.cs_subsys, item);
 761 }
 762
 763 int o2nm_depend_this_node(void)
 764 {
 765         int ret = 0;
 766         struct o2nm_node *local_node;
 767
 768         local_node = o2nm_get_node_by_num(o2nm_this_node());
 769         if (!local_node) {
 770                 ret = -EINVAL;
 771                 goto out;
 772         }
 773
 774         ret = o2nm_depend_item(&local_node->nd_item);
 775         o2nm_node_put(local_node);
 776
 777 out:
 778         return ret;
 779 }
 780
 781 void o2nm_undepend_this_node(void)
 782 {
 783         struct o2nm_node *local_node;
 784
 785         local_node = o2nm_get_node_by_num(o2nm_this_node());
 786         BUG_ON(!local_node);
 787
 788         o2nm_undepend_item(&local_node->nd_item);
 789         o2nm_node_put(local_node);
 790 }
 791
 792
 793 static void __exit exit_o2nm(void)
 794 {
 795         /* XXX sync with hb callbacks and shut down hb? */
 796         o2net_unregister_hb_callbacks();
 797         configfs_unregister_subsystem(&o2nm_cluster_group.cs_subsys);
 798         o2cb_sys_shutdown();
 799
 800         o2net_exit();
 801         o2hb_exit();
 802 }
 803
 804 static int __init init_o2nm(void)
 805 {
 806         int ret = -1;
 807
 808         ret = o2hb_init();
 809         if (ret)
 810                 goto out;
 811
 812         ret = o2net_init();
 813         if (ret)
 814                 goto out_o2hb;
 815
 816         ret = o2net_register_hb_callbacks();
 817         if (ret)
 818                 goto out_o2net;
 819
 820         config_group_init(&o2nm_cluster_group.cs_subsys.su_group);
 821         mutex_init(&o2nm_cluster_group.cs_subsys.su_mutex);
 822         ret = configfs_register_subsystem(&o2nm_cluster_group.cs_subsys);
 823         if (ret) {
 824                 printk(KERN_ERR "nodemanager: Registration returned %d\n", ret);
 825                 goto out_callbacks;
 826         }
 827
 828         ret = o2cb_sys_init();
 829         if (!ret)
 830                 goto out;
 831
 832         configfs_unregister_subsystem(&o2nm_cluster_group.cs_subsys);
 833 out_callbacks:
 834         o2net_unregister_hb_callbacks();
 835 out_o2net:
 836         o2net_exit();
 837 out_o2hb:
 838         o2hb_exit();
 839 out:
 840         return ret;
 841 }
 842
 843 MODULE_AUTHOR("Oracle");
 844 MODULE_LICENSE("GPL");
 845 MODULE_DESCRIPTION("OCFS2 cluster management");
 846
 847 module_init(init_o2nm)
 848 module_exit(exit_o2nm)